In this paper, I propose a test for a publication bias suggested by Viscusi and Masterman [2017]. This bias arises from the selection of estimates based on their distance from the usual range in the literature. It implies that, in a given research area, there is a progressive concentration of estimates around this range as the literature grows. I test this hypothesis using OECD meta-data on VSL elicited via stated preferences. Using interquantile regression, I am able to model semi-parametrically the dispersion of estimates as a function of the number of prior studies, controlling for confounding factors. Consistently with the hypothesis, my results show a decrease in the dispersion for the lower part of the conditional distribution. I find no significant pattern for the upper part of the conditional distribution. This result has two implications. First, there is a systematic underestimation of the variance of mean estimates in meta-analyses of VSL. Second, there is an upward bias in the mean estimate of VSL meta-analyses due to the asymmetry of the selection pattern. Both implications might also affect meta-analyses in other research areas.